Expression body indent wrong

[JakeWharton]

Today:

  public override val descriptor: SerialDescriptor = buildString {
      }

  public override val descriptor: SerialDescriptor =
      buildClassSerialDescriptor("RowVerticalAlignment") {
      }

Should be:

  public override val descriptor: SerialDescriptor = buildString {
  }

  public override val descriptor: SerialDescriptor =
    buildClassSerialDescriptor("RowVerticalAlignment") {
    }

We need to remove the function indent when it collapses so that wraps are +2 and begin/end control flows can have their curly's flush with the function's visibility modifier indent level.

[JakeWharton]

I made the second example a non-empty block and things got weirder:

  public override val descriptor: SerialDescriptor =
      buildClassSerialDescriptor("RowVerticalAlignment") {
    element<Int>("direction")
  }

Here's the code:

addProperty(PropertySpec.builder("descriptor", SerialDescriptor)
  .addModifiers(OVERRIDE)
  .initializer(CodeBlock.builder()
    .beginControlFlow("%M(%S)", buildClassSerialDescriptor, layoutModifierType.simpleName)
    .apply {
      for (property in layoutModifier.traits) {
        addStatement("%M<%T>(%S)", element, property.type.asTypeName(), property.name)
      }
    }
    .endControlFlow()
    .build())
  .build())

[JakeWharton]

  public override fun equals(other: Any?): Boolean = other is AccessibilityDescription
  && other.`value` == `value`

Beginning to think expression body support was a mistake because it seems to be wrong more often than not.

[Egorand]

I dug a bit into how we emit PropertySpec.initializer, and the logic is that we'll wrap the entire CodeBlock into a statement («%L») unless it already includes statements. Emitting the CodeBlock as a single statement seems to produce expected results. Maybe we should just do a better job processing the CodeBlock before emitting it?